1 /* Copyright (c) 2008-2014. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
10 #define DW_LANG_Objc DW_LANG_ObjC /* fix spelling error in older dwarf.h */
12 #include <elfutils/libdw.h>
14 #include <simgrid_config.h>
16 #include <xbt/sysdep.h>
18 #include "mc_object_info.h"
19 #include "mc_private.h"
21 static void mc_variable_free(mc_variable_t v);
22 static void mc_variable_free_voidp(void *t);
24 static void MC_dwarf_register_global_variable(mc_object_info_t info, mc_variable_t variable);
25 static void MC_register_variable(mc_object_info_t info, dw_frame_t frame, mc_variable_t variable);
26 static void MC_dwarf_register_non_global_variable(mc_object_info_t info, dw_frame_t frame, mc_variable_t variable);
27 static void MC_dwarf_register_variable(mc_object_info_t info, dw_frame_t frame, mc_variable_t variable);
29 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_dwarf, mc, "DWARF processing");
31 /** \brief The default DW_TAG_lower_bound for a given DW_AT_language.
33 * The default for a given language is defined in the DWARF spec.
35 * \param language consant as defined by the DWARf spec
37 static uint64_t MC_dwarf_default_lower_bound(int lang);
39 /** \brief Computes the the element_count of a DW_TAG_enumeration_type DIE
41 * This is the number of elements in a given array dimension.
43 * A reference of the compilation unit (DW_TAG_compile_unit) is
44 * needed because the default lower bound (when there is no DW_AT_lower_bound)
45 * depends of the language of the compilation unit (DW_AT_language).
47 * \param die DIE for the DW_TAG_enumeration_type or DW_TAG_subrange_type
48 * \param unit DIE of the DW_TAG_compile_unit
50 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die * die,
53 /** \brief Computes the number of elements of a given DW_TAG_array_type.
55 * \param die DIE for the DW_TAG_array_type
57 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit);
59 /** \brief Process a DIE
61 * \param info the resulting object fot the library/binary file (output)
62 * \param die the current DIE
63 * \param unit the DIE of the compile unit of the current DIE
64 * \param frame containg frame if any
66 static void MC_dwarf_handle_die(mc_object_info_t info, Dwarf_Die * die,
67 Dwarf_Die * unit, dw_frame_t frame,
70 /** \brief Process a type DIE
72 static void MC_dwarf_handle_type_die(mc_object_info_t info, Dwarf_Die * die,
73 Dwarf_Die * unit, dw_frame_t frame,
76 /** \brief Calls MC_dwarf_handle_die on all childrend of the given die
78 * \param info the resulting object fot the library/binary file (output)
79 * \param die the current DIE
80 * \param unit the DIE of the compile unit of the current DIE
81 * \param frame containg frame if any
83 static void MC_dwarf_handle_children(mc_object_info_t info, Dwarf_Die * die,
84 Dwarf_Die * unit, dw_frame_t frame,
87 /** \brief Handle a variable (DW_TAG_variable or other)
89 * \param info the resulting object fot the library/binary file (output)
90 * \param die the current DIE
91 * \param unit the DIE of the compile unit of the current DIE
92 * \param frame containg frame if any
94 static void MC_dwarf_handle_variable_die(mc_object_info_t info, Dwarf_Die * die,
95 Dwarf_Die * unit, dw_frame_t frame,
98 /** \brief Get the DW_TAG_type of the DIE
101 * \return DW_TAG_type attribute as a new string (NULL if none)
103 static std::string MC_dwarf_at_type(Dwarf_Die * die);
105 /** \brief A class of DWARF tags (DW_TAG_*)
107 typedef enum mc_tag_class {
116 static mc_tag_class MC_dwarf_tag_classify(int tag)
120 case DW_TAG_array_type:
121 case DW_TAG_class_type:
122 case DW_TAG_enumeration_type:
124 case DW_TAG_pointer_type:
125 case DW_TAG_reference_type:
126 case DW_TAG_rvalue_reference_type:
127 case DW_TAG_string_type:
128 case DW_TAG_structure_type:
129 case DW_TAG_subroutine_type:
130 case DW_TAG_union_type:
131 case DW_TAG_ptr_to_member_type:
132 case DW_TAG_set_type:
133 case DW_TAG_subrange_type:
134 case DW_TAG_base_type:
135 case DW_TAG_const_type:
136 case DW_TAG_file_type:
137 case DW_TAG_packed_type:
138 case DW_TAG_volatile_type:
139 case DW_TAG_restrict_type:
140 case DW_TAG_interface_type:
141 case DW_TAG_unspecified_type:
142 case DW_TAG_shared_type:
145 case DW_TAG_subprogram:
146 return mc_tag_subprogram;
148 case DW_TAG_variable:
149 case DW_TAG_formal_parameter:
150 return mc_tag_variable;
152 case DW_TAG_lexical_block:
153 case DW_TAG_try_block:
154 case DW_TAG_catch_block:
155 case DW_TAG_inlined_subroutine:
156 case DW_TAG_with_stmt:
159 case DW_TAG_namespace:
160 return mc_tag_namespace;
163 return mc_tag_unknown;
168 #define MC_DW_CLASS_UNKNOWN 0
169 #define MC_DW_CLASS_ADDRESS 1 // Location in the address space of the program
170 #define MC_DW_CLASS_BLOCK 2 // Arbitrary block of bytes
171 #define MC_DW_CLASS_CONSTANT 3
172 #define MC_DW_CLASS_STRING 3 // String
173 #define MC_DW_CLASS_FLAG 4 // Boolean
174 #define MC_DW_CLASS_REFERENCE 5 // Reference to another DIE
175 #define MC_DW_CLASS_EXPRLOC 6 // DWARF expression/location description
176 #define MC_DW_CLASS_LINEPTR 7
177 #define MC_DW_CLASS_LOCLISTPTR 8
178 #define MC_DW_CLASS_MACPTR 9
179 #define MC_DW_CLASS_RANGELISTPTR 10
181 /** \brief Find the DWARF data class for a given DWARF data form
183 * This mapping is defined in the DWARF spec.
185 * \param form The form (values taken from the DWARF spec)
186 * \return An internal representation for the corresponding class
188 static int MC_dwarf_form_get_class(int form)
192 return MC_DW_CLASS_ADDRESS;
197 return MC_DW_CLASS_BLOCK;
204 return MC_DW_CLASS_CONSTANT;
207 return MC_DW_CLASS_STRING;
208 case DW_FORM_ref_addr:
213 case DW_FORM_ref_udata:
214 return MC_DW_CLASS_REFERENCE;
216 case DW_FORM_flag_present:
217 return MC_DW_CLASS_FLAG;
218 case DW_FORM_exprloc:
219 return MC_DW_CLASS_EXPRLOC;
223 return MC_DW_CLASS_UNKNOWN;
227 /** \brief Get the name of the tag of a given DIE
230 * \return name of the tag of this DIE
232 static inline const char *MC_dwarf_die_tagname(Dwarf_Die * die)
234 return MC_dwarf_tagname(dwarf_tag(die));
239 /** \brief Get an attribute of a given DIE as a string
242 * \param attribute attribute
243 * \return value of the given attribute of the given DIE
245 static const char *MC_dwarf_attr_integrate_string(Dwarf_Die * die,
248 Dwarf_Attribute attr;
249 if (!dwarf_attr_integrate(die, attribute, &attr)) {
252 return dwarf_formstring(&attr);
256 /** \brief Get the linkage name of a DIE.
258 * Use either DW_AT_linkage_name or DW_AT_MIPS_linkage_name.
259 * DW_AT_linkage_name is standardized since DWARF 4.
260 * Before this version of DWARF, the MIPS extensions
261 * DW_AT_MIPS_linkage_name is used (at least by GCC).
264 * \return linkage name of the given DIE (or NULL)
266 static const char *MC_dwarf_at_linkage_name(Dwarf_Die * die)
268 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_linkage_name);
270 name = MC_dwarf_attr_integrate_string(die, DW_AT_MIPS_linkage_name);
274 static Dwarf_Off MC_dwarf_attr_dieoffset(Dwarf_Die * die, int attribute)
276 Dwarf_Attribute attr;
277 if (dwarf_hasattr_integrate(die, attribute)) {
278 dwarf_attr_integrate(die, attribute, &attr);
279 Dwarf_Die subtype_die;
280 if (dwarf_formref_die(&attr, &subtype_die) == NULL) {
281 xbt_die("Could not find DIE");
283 return dwarf_dieoffset(&subtype_die);
288 static Dwarf_Off MC_dwarf_attr_integrate_dieoffset(Dwarf_Die * die,
291 Dwarf_Attribute attr;
292 if (dwarf_hasattr_integrate(die, attribute)) {
293 dwarf_attr_integrate(die, DW_AT_type, &attr);
294 Dwarf_Die subtype_die;
295 if (dwarf_formref_die(&attr, &subtype_die) == NULL) {
296 xbt_die("Could not find DIE");
298 return dwarf_dieoffset(&subtype_die);
303 /** \brief Find the type/subtype (DW_AT_type) for a DIE
306 * \return DW_AT_type reference as a global offset in hexadecimal (or NULL)
309 std::string MC_dwarf_at_type(Dwarf_Die * die)
311 Dwarf_Off offset = MC_dwarf_attr_integrate_dieoffset(die, DW_AT_type);
313 return std::string();
314 char* s = bprintf("%" PRIx64, offset);
317 return std::move(res);
320 static uint64_t MC_dwarf_attr_integrate_addr(Dwarf_Die * die, int attribute)
322 Dwarf_Attribute attr;
323 if (dwarf_attr_integrate(die, attribute, &attr) == NULL)
326 if (dwarf_formaddr(&attr, &value) == 0)
327 return (uint64_t) value;
332 static uint64_t MC_dwarf_attr_integrate_uint(Dwarf_Die * die, int attribute,
333 uint64_t default_value)
335 Dwarf_Attribute attr;
336 if (dwarf_attr_integrate(die, attribute, &attr) == NULL)
337 return default_value;
339 return dwarf_formudata(dwarf_attr_integrate(die, attribute, &attr),
340 &value) == 0 ? (uint64_t) value : default_value;
343 static bool MC_dwarf_attr_flag(Dwarf_Die * die, int attribute, bool integrate)
345 Dwarf_Attribute attr;
346 if ((integrate ? dwarf_attr_integrate(die, attribute, &attr)
347 : dwarf_attr(die, attribute, &attr)) == 0)
351 if (dwarf_formflag(&attr, &result))
352 xbt_die("Unexpected form for attribute %s", MC_dwarf_attrname(attribute));
356 /** \brief Find the default lower bound for a given language
358 * The default lower bound of an array (when DW_TAG_lower_bound
359 * is missing) depends on the language of the compilation unit.
361 * \param lang Language of the compilation unit (values defined in the DWARF spec)
362 * \return Default lower bound of an array in this compilation unit
364 static uint64_t MC_dwarf_default_lower_bound(int lang)
370 case DW_LANG_C_plus_plus:
374 case DW_LANG_ObjC_plus_plus:
380 case DW_LANG_Fortran77:
381 case DW_LANG_Fortran90:
382 case DW_LANG_Fortran95:
383 case DW_LANG_Modula2:
384 case DW_LANG_Pascal83:
386 case DW_LANG_Cobol74:
387 case DW_LANG_Cobol85:
390 xbt_die("No default DW_TAG_lower_bound for language %i and none given",
396 /** \brief Finds the number of elements in a DW_TAG_subrange_type or DW_TAG_enumeration_type DIE
399 * \param unit DIE of the compilation unit
400 * \return number of elements in the range
402 static uint64_t MC_dwarf_subrange_element_count(Dwarf_Die * die,
405 xbt_assert(dwarf_tag(die) == DW_TAG_enumeration_type
406 || dwarf_tag(die) == DW_TAG_subrange_type,
407 "MC_dwarf_subrange_element_count called with DIE of type %s",
408 MC_dwarf_die_tagname(die));
410 // Use DW_TAG_count if present:
411 if (dwarf_hasattr_integrate(die, DW_AT_count)) {
412 return MC_dwarf_attr_integrate_uint(die, DW_AT_count, 0);
414 // Otherwise compute DW_TAG_upper_bound-DW_TAG_lower_bound + 1:
416 if (!dwarf_hasattr_integrate(die, DW_AT_upper_bound)) {
417 // This is not really 0, but the code expects this (we do not know):
420 uint64_t upper_bound =
421 MC_dwarf_attr_integrate_uint(die, DW_AT_upper_bound, -1);
423 uint64_t lower_bound = 0;
424 if (dwarf_hasattr_integrate(die, DW_AT_lower_bound)) {
425 lower_bound = MC_dwarf_attr_integrate_uint(die, DW_AT_lower_bound, -1);
427 lower_bound = MC_dwarf_default_lower_bound(dwarf_srclang(unit));
429 return upper_bound - lower_bound + 1;
432 /** \brief Finds the number of elements in a array type (DW_TAG_array_type)
434 * The compilation unit might be needed because the default lower
435 * bound depends on the language of the compilation unit.
437 * \param die the DIE of the DW_TAG_array_type
438 * \param unit the DIE of the compilation unit
439 * \return number of elements in this array type
441 static uint64_t MC_dwarf_array_element_count(Dwarf_Die * die, Dwarf_Die * unit)
443 xbt_assert(dwarf_tag(die) == DW_TAG_array_type,
444 "MC_dwarf_array_element_count called with DIE of type %s",
445 MC_dwarf_die_tagname(die));
450 for (res = dwarf_child(die, &child); res == 0;
451 res = dwarf_siblingof(&child, &child)) {
452 int child_tag = dwarf_tag(&child);
453 if (child_tag == DW_TAG_subrange_type
454 || child_tag == DW_TAG_enumeration_type) {
455 result *= MC_dwarf_subrange_element_count(&child, unit);
463 /** \brief Initialize the location of a member of a type
464 * (DW_AT_data_member_location of a DW_TAG_member).
466 * \param type a type (struct, class)
467 * \param member the member of the type
468 * \param child DIE of the member (DW_TAG_member)
470 static void MC_dwarf_fill_member_location(mc_type_t type, mc_type_t member,
473 if (dwarf_hasattr(child, DW_AT_data_bit_offset)) {
474 xbt_die("Can't groke DW_AT_data_bit_offset.");
477 if (!dwarf_hasattr_integrate(child, DW_AT_data_member_location)) {
478 if (type->type != DW_TAG_union_type) {
480 ("Missing DW_AT_data_member_location field in DW_TAG_member %s of type <%"
481 PRIx64 ">%s", member->name.c_str(),
482 (uint64_t) type->id, type->name.c_str());
488 Dwarf_Attribute attr;
489 dwarf_attr_integrate(child, DW_AT_data_member_location, &attr);
490 int form = dwarf_whatform(&attr);
491 int klass = MC_dwarf_form_get_class(form);
493 case MC_DW_CLASS_EXPRLOC:
494 case MC_DW_CLASS_BLOCK:
495 // Location expression:
499 if (dwarf_getlocation(&attr, &expr, &len)) {
501 ("Could not read location expression DW_AT_data_member_location in DW_TAG_member %s of type <%"
502 PRIx64 ">%s", MC_dwarf_attr_integrate_string(child, DW_AT_name),
503 (uint64_t) type->id, type->name.c_str());
505 simgrid::mc::DwarfExpression(expr, expr+len);
508 case MC_DW_CLASS_CONSTANT:
509 // Offset from the base address of the object:
512 if (!dwarf_formudata(&attr, &offset))
513 member->offset(offset);
515 xbt_die("Cannot get %s location <%" PRIx64 ">%s",
516 MC_dwarf_attr_integrate_string(child, DW_AT_name),
517 (uint64_t) type->id, type->name.c_str());
520 case MC_DW_CLASS_LOCLISTPTR:
521 // Reference to a location list:
523 case MC_DW_CLASS_REFERENCE:
524 // It's supposed to be possible in DWARF2 but I couldn't find its semantic
527 xbt_die("Can't handle form class (%i) / form 0x%x as DW_AT_member_location",
533 static void dw_type_free_voidp(void *t)
535 delete *(mc_type_t*)t;
538 /** \brief Populate the list of members of a type
540 * \param info ELF object containing the type DIE
541 * \param die DIE of the type
542 * \param unit DIE of the compilation unit containing the type DIE
543 * \param type the type
545 static void MC_dwarf_add_members(mc_object_info_t info, Dwarf_Die * die,
546 Dwarf_Die * unit, mc_type_t type)
550 xbt_assert(type->members.empty());
551 for (res = dwarf_child(die, &child); res == 0;
552 res = dwarf_siblingof(&child, &child)) {
553 int tag = dwarf_tag(&child);
554 if (tag == DW_TAG_member || tag == DW_TAG_inheritance) {
556 // Skip declarations:
557 if (MC_dwarf_attr_flag(&child, DW_AT_declaration, false))
560 // Skip compile time constants:
561 if (dwarf_hasattr(&child, DW_AT_const_value))
564 // TODO, we should use another type (because is is not a type but a member)
565 simgrid::mc::Type member;
569 member.id = dwarf_dieoffset(&child);
571 const char *name = MC_dwarf_attr_integrate_string(&child, DW_AT_name);
575 MC_dwarf_attr_integrate_uint(&child, DW_AT_byte_size, 0);
576 member.element_count = -1;
577 member.type_id = MC_dwarf_at_type(&child);
579 if (dwarf_hasattr(&child, DW_AT_data_bit_offset)) {
580 xbt_die("Can't groke DW_AT_data_bit_offset.");
583 MC_dwarf_fill_member_location(type, &member, &child);
585 if (member.type_id.empty()) {
586 xbt_die("Missing type for member %s of <%" PRIx64 ">%s",
588 (uint64_t) type->id, type->name.c_str());
591 type->members.push_back(std::move(member));
596 /** \brief Create a MC type object from a DIE
598 * \param info current object info object
599 * \param DIE (for a given type);
600 * \param unit compilation unit of the current DIE
601 * \return MC representation of the type
603 static mc_type_t MC_dwarf_die_to_type(mc_object_info_t info, Dwarf_Die * die,
604 Dwarf_Die * unit, dw_frame_t frame,
608 mc_type_t type = new simgrid::mc::Type();
610 type->name = std::string();
611 type->element_count = -1;
613 type->type = dwarf_tag(die);
616 type->id = dwarf_dieoffset(die);
618 const char *prefix = "";
619 switch (type->type) {
620 case DW_TAG_structure_type:
623 case DW_TAG_union_type:
626 case DW_TAG_class_type:
633 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
635 char* full_name = ns ? bprintf("%s%s::%s", prefix, ns, name) :
636 bprintf("%s%s", prefix, name);
637 type->name = std::string(full_name);
641 type->type_id = MC_dwarf_at_type(die);
643 // Some compilers do not emit DW_AT_byte_size for pointer_type,
644 // so we fill this. We currently assume that the model-checked process is in
645 // the same architecture..
646 if (type->type == DW_TAG_pointer_type)
647 type->byte_size = sizeof(void*);
649 // Computation of the byte_size;
650 if (dwarf_hasattr_integrate(die, DW_AT_byte_size))
651 type->byte_size = MC_dwarf_attr_integrate_uint(die, DW_AT_byte_size, 0);
652 else if (type->type == DW_TAG_array_type
653 || type->type == DW_TAG_structure_type
654 || type->type == DW_TAG_class_type) {
656 if (dwarf_aggregate_size(die, &size) == 0) {
657 type->byte_size = size;
661 switch (type->type) {
662 case DW_TAG_array_type:
663 type->element_count = MC_dwarf_array_element_count(die, unit);
664 // TODO, handle DW_byte_stride and (not) DW_bit_stride
667 case DW_TAG_pointer_type:
668 case DW_TAG_reference_type:
669 case DW_TAG_rvalue_reference_type:
670 type->is_pointer_type = 1;
673 case DW_TAG_structure_type:
674 case DW_TAG_union_type:
675 case DW_TAG_class_type:
676 MC_dwarf_add_members(info, die, unit, type);
677 char *new_ns = ns == NULL ? xbt_strdup(type->name.c_str())
678 : bprintf("%s::%s", ns, name);
679 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
687 static void MC_dwarf_handle_type_die(mc_object_info_t info, Dwarf_Die * die,
688 Dwarf_Die * unit, dw_frame_t frame,
691 mc_type_t type = MC_dwarf_die_to_type(info, die, unit, frame, ns);
693 char *key = bprintf("%" PRIx64, (uint64_t) type->id);
694 xbt_dict_set(info->types, key, type, NULL);
697 if (!type->name.empty() && type->byte_size != 0) {
698 xbt_dict_set(info->full_types_by_name, type->name.c_str(), type, NULL);
702 static int mc_anonymous_variable_index = 0;
704 static mc_variable_t MC_die_to_variable(mc_object_info_t info, Dwarf_Die * die,
705 Dwarf_Die * unit, dw_frame_t frame,
708 // Skip declarations:
709 if (MC_dwarf_attr_flag(die, DW_AT_declaration, false))
712 // Skip compile time constants:
713 if (dwarf_hasattr(die, DW_AT_const_value))
716 Dwarf_Attribute attr_location;
717 if (dwarf_attr(die, DW_AT_location, &attr_location) == NULL) {
718 // No location: do not add it ?
722 simgrid::mc::Variable* variable = new simgrid::mc::Variable();
723 variable->dwarf_offset = dwarf_dieoffset(die);
724 variable->global = frame == NULL; // Can be override base on DW_AT_location
725 variable->object_info = info;
727 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
729 variable->name = name;
730 variable->type_id = MC_dwarf_at_type(die);
732 int form = dwarf_whatform(&attr_location);
735 DW_FORM_sec_offset ? MC_DW_CLASS_CONSTANT : MC_dwarf_form_get_class(form);
737 case MC_DW_CLASS_EXPRLOC:
738 case MC_DW_CLASS_BLOCK:
739 // Location expression:
743 if (dwarf_getlocation(&attr_location, &expr, &len)) {
745 "Could not read location expression in DW_AT_location "
746 "of variable <%" PRIx64 ">%s",
747 (uint64_t) variable->dwarf_offset,
748 variable->name.c_str());
751 if (len == 1 && expr[0].atom == DW_OP_addr) {
752 variable->global = 1;
753 uintptr_t offset = (uintptr_t) expr[0].number;
754 uintptr_t base = (uintptr_t) info->base_address();
755 variable->address = (void *) (base + offset);
757 mc_dwarf_location_list_init_from_expression(
758 &variable->location_list, len,
764 case MC_DW_CLASS_LOCLISTPTR:
765 case MC_DW_CLASS_CONSTANT:
766 // Reference to location list:
767 mc_dwarf_location_list_init(
768 &variable->location_list, info, die,
772 xbt_die("Unexpected form 0x%x (%i), class 0x%x (%i) list for location "
773 "in <%" PRIx64 ">%s",
774 form, form, klass, klass,
775 (uint64_t) variable->dwarf_offset,
776 variable->name.c_str());
779 // Handle start_scope:
780 if (dwarf_hasattr(die, DW_AT_start_scope)) {
781 Dwarf_Attribute attr;
782 dwarf_attr(die, DW_AT_start_scope, &attr);
783 int form = dwarf_whatform(&attr);
784 int klass = MC_dwarf_form_get_class(form);
786 case MC_DW_CLASS_CONSTANT:
789 variable->start_scope =
790 dwarf_formudata(&attr, &value) == 0 ? (size_t) value : 0;
793 case MC_DW_CLASS_RANGELISTPTR: // TODO
796 ("Unhandled form 0x%x, class 0x%X for DW_AT_start_scope of variable %s",
797 form, klass, name == NULL ? "?" : name);
801 if (ns && variable->global)
803 std::string(ns) + "::" + variable->name;
805 // The current code needs a variable name,
806 // generate a fake one:
807 if (variable->name.empty())
809 "@anonymous#" + std::to_string(mc_anonymous_variable_index++);
814 static void MC_dwarf_handle_variable_die(mc_object_info_t info, Dwarf_Die * die,
815 Dwarf_Die * unit, dw_frame_t frame,
818 mc_variable_t variable =
819 MC_die_to_variable(info, die, unit, frame, ns);
820 if (variable == NULL)
822 MC_dwarf_register_variable(info, frame, variable);
825 static void mc_frame_free_voipd(dw_frame_t * p)
831 static void MC_dwarf_handle_scope_die(mc_object_info_t info, Dwarf_Die * die,
832 Dwarf_Die * unit, dw_frame_t parent_frame,
835 // TODO, handle DW_TAG_type/DW_TAG_location for DW_TAG_with_stmt
836 int tag = dwarf_tag(die);
837 mc_tag_class klass = MC_dwarf_tag_classify(tag);
839 // (Template) Subprogram declaration:
840 if (klass == mc_tag_subprogram
841 && MC_dwarf_attr_flag(die, DW_AT_declaration, false))
844 if (klass == mc_tag_scope)
845 xbt_assert(parent_frame, "No parent scope for this scope");
847 dw_frame_t frame = xbt_new0(s_dw_frame_t, 1);
850 frame->id = dwarf_dieoffset(die);
851 frame->object_info = info;
853 if (klass == mc_tag_subprogram) {
854 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
856 ns ? bprintf("%s::%s", ns, name) : xbt_strdup(name);
859 frame->abstract_origin_id =
860 MC_dwarf_attr_dieoffset(die, DW_AT_abstract_origin);
862 // This is the base address for DWARF addresses.
863 // Relocated addresses are offset from this base address.
864 // See DWARF4 spec 7.5
865 void *base = info->base_address();
867 // Variables are filled in the (recursive) call of MC_dwarf_handle_children:
869 xbt_dynar_new(sizeof(mc_variable_t), mc_variable_free_voidp);
871 // TODO, support DW_AT_ranges
872 uint64_t low_pc = MC_dwarf_attr_integrate_addr(die, DW_AT_low_pc);
873 frame->low_pc = low_pc ? ((char *) base) + low_pc : 0;
876 Dwarf_Attribute attr;
877 if (!dwarf_attr_integrate(die, DW_AT_high_pc, &attr)) {
878 xbt_die("Missing DW_AT_high_pc matching with DW_AT_low_pc");
884 switch (MC_dwarf_form_get_class(dwarf_whatform(&attr))) {
886 // DW_AT_high_pc if an offset from the low_pc:
887 case MC_DW_CLASS_CONSTANT:
889 if (dwarf_formsdata(&attr, &offset) != 0)
890 xbt_die("Could not read constant");
891 frame->high_pc = (void *) ((char *) frame->low_pc + offset);
894 // DW_AT_high_pc is a relocatable address:
895 case MC_DW_CLASS_ADDRESS:
896 if (dwarf_formaddr(&attr, &high_pc) != 0)
897 xbt_die("Could not read address");
898 frame->high_pc = ((char *) base) + high_pc;
902 xbt_die("Unexpected class for DW_AT_high_pc");
907 if (klass == mc_tag_subprogram) {
908 Dwarf_Attribute attr_frame_base;
909 if (dwarf_attr_integrate(die, DW_AT_frame_base, &attr_frame_base))
910 mc_dwarf_location_list_init(&frame->frame_base, info, die,
915 xbt_dynar_new(sizeof(dw_frame_t), (void_f_pvoid_t) mc_frame_free_voipd);
918 if (klass == mc_tag_subprogram) {
919 char *key = bprintf("%" PRIx64, (uint64_t) frame->id);
920 xbt_dict_set(info->subprograms, key, frame, NULL);
922 } else if (klass == mc_tag_scope) {
923 xbt_dynar_push(parent_frame->scopes, &frame);
926 MC_dwarf_handle_children(info, die, unit, frame, ns);
929 static void mc_dwarf_handle_namespace_die(mc_object_info_t info,
930 Dwarf_Die * die, Dwarf_Die * unit,
934 const char *name = MC_dwarf_attr_integrate_string(die, DW_AT_name);
936 xbt_die("Unexpected namespace in a subprogram");
937 char *new_ns = ns == NULL ? xbt_strdup(name)
938 : bprintf("%s::%s", ns, name);
939 MC_dwarf_handle_children(info, die, unit, frame, new_ns);
943 static void MC_dwarf_handle_children(mc_object_info_t info, Dwarf_Die * die,
944 Dwarf_Die * unit, dw_frame_t frame,
947 // For each child DIE:
950 for (res = dwarf_child(die, &child); res == 0;
951 res = dwarf_siblingof(&child, &child)) {
952 MC_dwarf_handle_die(info, &child, unit, frame, ns);
956 static void MC_dwarf_handle_die(mc_object_info_t info, Dwarf_Die * die,
957 Dwarf_Die * unit, dw_frame_t frame,
960 int tag = dwarf_tag(die);
961 mc_tag_class klass = MC_dwarf_tag_classify(tag);
966 MC_dwarf_handle_type_die(info, die, unit, frame, ns);
969 // Subprogram or scope:
970 case mc_tag_subprogram:
972 MC_dwarf_handle_scope_die(info, die, unit, frame, ns);
976 case mc_tag_variable:
977 MC_dwarf_handle_variable_die(info, die, unit, frame, ns);
980 case mc_tag_namespace:
981 mc_dwarf_handle_namespace_die(info, die, unit, frame, ns);
990 /** \brief Populate the debugging informations of the given ELF object
992 * Read the DWARf information of the EFFL object and populate the
993 * lists of types, variables, functions.
995 void MC_dwarf_get_variables(mc_object_info_t info)
997 int fd = open(info->file_name, O_RDONLY);
999 xbt_die("Could not open file %s", info->file_name);
1001 Dwarf *dwarf = dwarf_begin(fd, DWARF_C_READ);
1002 if (dwarf == NULL) {
1003 xbt_die("Your program must be compiled with -g (%s)", info->file_name);
1005 // For each compilation unit:
1006 Dwarf_Off offset = 0;
1007 Dwarf_Off next_offset = 0;
1009 while (dwarf_nextcu(dwarf, offset, &next_offset, &length, NULL, NULL, NULL) ==
1012 if (dwarf_offdie(dwarf, offset + length, &unit_die) != NULL) {
1014 // For each child DIE:
1017 for (res = dwarf_child(&unit_die, &child); res == 0;
1018 res = dwarf_siblingof(&child, &child)) {
1019 MC_dwarf_handle_die(info, &child, &unit_die, NULL, NULL);
1023 offset = next_offset;
1030 /************************** Free functions *************************/
1032 void mc_frame_free(dw_frame_t frame)
1034 xbt_free(frame->name);
1035 mc_dwarf_location_list_clear(&(frame->frame_base));
1036 xbt_dynar_free(&(frame->variables));
1037 xbt_dynar_free(&(frame->scopes));
1041 static void dw_type_free(mc_type_t t)
1046 void mc_variable_free_voidp(void *t)
1048 delete *(simgrid::mc::Variable**)t;
1051 // ***** object_info
1056 ObjectInformation::ObjectInformation()
1059 this->file_name = nullptr;
1060 this->start = nullptr;
1061 this->end = nullptr;
1062 this->start_exec = nullptr;
1063 this->end_exec = nullptr;
1064 this->start_rw = nullptr;
1065 this->end_rw = nullptr;
1066 this->start_ro = nullptr;
1067 this->end_ro = nullptr;
1068 this->subprograms = xbt_dict_new_homogeneous((void (*)(void *)) mc_frame_free);
1069 this->global_variables =
1070 xbt_dynar_new(sizeof(mc_variable_t), mc_variable_free_voidp);
1071 this->types = xbt_dict_new_homogeneous((void (*)(void *)) dw_type_free);
1072 this->full_types_by_name = xbt_dict_new_homogeneous(NULL);
1073 this->functions_index = nullptr;
1076 ObjectInformation::~ObjectInformation()
1078 xbt_free(this->file_name);
1079 xbt_dict_free(&this->subprograms);
1080 xbt_dynar_free(&this->global_variables);
1081 xbt_dict_free(&this->types);
1082 xbt_dict_free(&this->full_types_by_name);
1083 xbt_dynar_free(&this->functions_index);
1086 /** Find the DWARF offset for this ELF object
1088 * An offset is applied to address found in DWARF:
1091 * <li>for an executable obejct, addresses are virtual address
1092 * (there is no offset) i.e. \f$\text{virtual address} = \{dwarf address}\f$;</li>
1093 * <li>for a shared object, the addreses are offset from the begining
1094 * of the shared object (the base address of the mapped shared
1095 * object must be used as offset
1096 * i.e. \f$\text{virtual address} = \text{shared object base address}
1097 * + \text{dwarf address}\f$.</li>
1100 void *ObjectInformation::base_address() const
1102 if (this->executable())
1105 void *result = this->start_exec;
1106 if (this->start_rw != NULL && result > (void *) this->start_rw)
1107 result = this->start_rw;
1108 if (this->start_ro != NULL && result > (void *) this->start_ro)
1109 result = this->start_ro;
1118 // ***** Functions index
1120 static int MC_compare_frame_index_items(mc_function_index_item_t a,
1121 mc_function_index_item_t b)
1123 if (a->low_pc < b->low_pc)
1125 else if (a->low_pc == b->low_pc)
1131 static void MC_make_functions_index(mc_object_info_t info)
1133 xbt_dynar_t index = xbt_dynar_new(sizeof(s_mc_function_index_item_t), NULL);
1135 // Populate the array:
1136 dw_frame_t frame = NULL;
1137 xbt_dict_cursor_t cursor;
1139 xbt_dict_foreach(info->subprograms, cursor, key, frame) {
1140 if (frame->low_pc == NULL)
1142 s_mc_function_index_item_t entry;
1143 entry.low_pc = frame->low_pc;
1144 entry.high_pc = frame->high_pc;
1145 entry.function = frame;
1146 xbt_dynar_push(index, &entry);
1149 mc_function_index_item_t base =
1150 (mc_function_index_item_t) xbt_dynar_get_ptr(index, 0);
1152 // Sort the array by low_pc:
1154 xbt_dynar_length(index),
1155 sizeof(s_mc_function_index_item_t),
1156 (int (*)(const void *, const void *)) MC_compare_frame_index_items);
1158 info->functions_index = index;
1161 static void MC_post_process_variables(mc_object_info_t info)
1163 unsigned cursor = 0;
1164 mc_variable_t variable = nullptr;
1165 xbt_dynar_foreach(info->global_variables, cursor, variable)
1166 if (!variable->type_id.empty())
1167 variable->type = (mc_type_t) xbt_dict_get_or_null(
1168 info->types, variable->type_id.c_str());
1171 static void mc_post_process_scope(mc_object_info_t info, dw_frame_t scope)
1174 if (scope->tag == DW_TAG_inlined_subroutine) {
1176 // Attach correct namespaced name in inlined subroutine:
1177 char *key = bprintf("%" PRIx64, (uint64_t) scope->abstract_origin_id);
1178 dw_frame_t abstract_origin = (dw_frame_t) xbt_dict_get_or_null(info->subprograms, key);
1179 xbt_assert(abstract_origin, "Could not lookup abstract origin %s", key);
1181 scope->name = xbt_strdup(abstract_origin->name);
1186 unsigned cursor = 0;
1187 mc_variable_t variable = nullptr;
1188 xbt_dynar_foreach(scope->variables, cursor, variable)
1189 if (!variable->type_id.empty())
1190 variable->type = (mc_type_t) xbt_dict_get_or_null(
1191 info->types, variable->type_id.c_str());
1193 // Recursive post-processing of nested-scopes:
1194 dw_frame_t nested_scope = nullptr;
1195 xbt_dynar_foreach(scope->scopes, cursor, nested_scope)
1196 mc_post_process_scope(info, nested_scope);
1200 static void MC_post_process_functions(mc_object_info_t info)
1202 xbt_dict_cursor_t cursor;
1204 dw_frame_t subprogram = NULL;
1205 xbt_dict_foreach(info->subprograms, cursor, key, subprogram) {
1206 mc_post_process_scope(info, subprogram);
1211 /** \brief Fill/lookup the "subtype" field.
1213 static void MC_resolve_subtype(mc_object_info_t info, mc_type_t type)
1216 if (type->type_id.empty())
1218 type->subtype = (mc_type_t) xbt_dict_get_or_null(
1219 info->types, type->type_id.c_str());
1220 if (type->subtype == NULL)
1222 if (type->subtype->byte_size != 0)
1224 if (type->subtype->name.empty())
1226 // Try to find a more complete description of the type:
1227 // We need to fix in order to support C++.
1230 (mc_type_t) xbt_dict_get_or_null(
1231 info->full_types_by_name, type->subtype->name.c_str());
1232 if (subtype != NULL) {
1233 type->subtype = subtype;
1238 static void MC_post_process_types(mc_object_info_t info)
1240 xbt_dict_cursor_t cursor = NULL;
1244 // Lookup "subtype" field:
1245 xbt_dict_foreach(info->types, cursor, origin, type) {
1246 MC_resolve_subtype(info, type);
1247 for (simgrid::mc::Type& member : type->members)
1248 MC_resolve_subtype(info, &member);
1252 /** \brief Finds informations about a given shared object/executable */
1253 std::shared_ptr<s_mc_object_info_t> MC_find_object_info(
1254 std::vector<simgrid::mc::VmMap> const& maps, const char *name, int executable)
1256 std::shared_ptr<s_mc_object_info_t> result =
1257 std::make_shared<s_mc_object_info_t>();
1259 result->flags |= MC_OBJECT_INFO_EXECUTABLE;
1260 result->file_name = xbt_strdup(name);
1261 MC_find_object_address(maps, result.get());
1262 MC_dwarf_get_variables(result.get());
1263 MC_post_process_types(result.get());
1264 MC_post_process_variables(result.get());
1265 MC_post_process_functions(result.get());
1266 MC_make_functions_index(result.get());
1267 return std::move(result);
1270 /*************************************************************************/
1272 static int MC_dwarf_get_variable_index(xbt_dynar_t variables, const char *var,
1276 if (xbt_dynar_is_empty(variables))
1279 unsigned int cursor = 0;
1281 int end = xbt_dynar_length(variables) - 1;
1282 mc_variable_t var_test = NULL;
1284 while (start <= end) {
1285 cursor = (start + end) / 2;
1287 (mc_variable_t) xbt_dynar_get_as(variables, cursor, mc_variable_t);
1288 if (strcmp(var_test->name.c_str(), var) < 0) {
1290 } else if (strcmp(var_test->name.c_str(), var) > 0) {
1293 if (address) { /* global variable */
1294 if (var_test->address == address)
1296 if (var_test->address > address)
1300 } else { /* local variable */
1306 if (strcmp(var_test->name.c_str(), var) == 0) {
1307 if (address && var_test->address < address)
1311 } else if (strcmp(var_test->name.c_str(), var) < 0)
1318 void MC_dwarf_register_global_variable(mc_object_info_t info,
1319 mc_variable_t variable)
1322 MC_dwarf_get_variable_index(info->global_variables,
1323 variable->name.c_str(),
1326 xbt_dynar_insert_at(info->global_variables, index, &variable);
1330 void MC_dwarf_register_non_global_variable(mc_object_info_t info,
1332 mc_variable_t variable)
1334 xbt_assert(frame, "Frame is NULL");
1336 MC_dwarf_get_variable_index(
1337 frame->variables, variable->name.c_str(), NULL);
1339 xbt_dynar_insert_at(frame->variables, index, &variable);
1343 void MC_dwarf_register_variable(mc_object_info_t info, dw_frame_t frame,
1344 mc_variable_t variable)
1346 if (variable->global)
1347 MC_dwarf_register_global_variable(info, variable);
1348 else if (frame == NULL)
1349 xbt_die("No frame for this local variable");
1351 MC_dwarf_register_non_global_variable(info, frame, variable);
1354 void MC_post_process_object_info(mc_process_t process, mc_object_info_t info)
1356 xbt_dict_cursor_t cursor = NULL;
1358 mc_type_t type = NULL;
1359 xbt_dict_foreach(info->types, cursor, key, type) {
1361 mc_type_t subtype = type;
1362 while (subtype->type == DW_TAG_typedef || subtype->type == DW_TAG_volatile_type
1363 || subtype->type == DW_TAG_const_type) {
1364 if (subtype->subtype)
1365 subtype = subtype->subtype;
1370 // Resolve full_type:
1371 if (!subtype->name.empty() && subtype->byte_size == 0) {
1372 for (auto const& object_info : process->object_infos) {
1373 mc_type_t same_type = (mc_type_t)
1374 xbt_dict_get_or_null(object_info->full_types_by_name,
1375 subtype->name.c_str());
1376 if (same_type && !same_type->name.empty() && same_type->byte_size) {
1377 type->full_type = same_type;
1381 } else type->full_type = subtype;